[Mlir-commits] [mlir] [mlir][vector] Canonicalize transfer_{read, write} vector<1xT> (PR #196598)

[Erick Ochoa Lopez]

amd-eochoalo wrote:

> "high-level vector dialect" and "low-level vector dialect". If this is something we really want, then we should formalize it and document the respective models, so that we can reason about lowering from high to low, and also reason about canonicalization and keeping it "closed" into these two "sub-dialects".

Thanks @mehdi and @dcaballe! Just pointing out that the MLIR vector dialect documentation already has a notion of "high-level vector dialect" (virtual vector) and "low-level vector dialect" (hardware vector), with vector.transfer_read included in the [list of "virtual vector ops"](https://mlir.llvm.org/docs/Dialects/Vector/#virtual-vector-ops).

I'm not sure how formal this separation needs to be and how that might impact the vector dialect, but happy to learn more about what that would look like and be part of the discussion. But that is probably beyond the scope of this PR and can be done as a follow up.

https://github.com/llvm/llvm-project/pull/196598